Researchers at Google DeepMind have introduced a new artificial intelligence model called AlphaGenome 2, which aims to reveal how genetic changes impact different cells. The system offers scientists a highly detailed perspective on a complex biological challenge, potentially uncovering crucial clues about disease origins and guiding the development of novel treatments.
Mapping Cellular Impact
AlphaGenome 2 builds upon DeepMind’s previous genome-predicting model, AlphaGenome, which launched the prior year. Although nearly all cells in the human body share identical DNA, different cell types activate distinct genes depending on their function and state. Consequently, a single genetic variant can produce vastly different outcomes depending on the specific cell type involved.
The updated model is built to capture these nuances by forecasting how genetic variants affect individual cell types and states. In a blog post published on Wednesday, Google explained that the system provides a much finer-grained perspective than broader models, which could uncover disease-related impacts in specific cells that might otherwise remain undetected.
Training and Capabilities
To achieve this level of detail, AlphaGenome 2 was trained on more than 100 million individual cell profiles spanning 68 tissues. Google DeepMind noted that the model effectively transforms static cell atlases into a predictive engine for human genetics. Furthermore, the company reported that AlphaGenome 2 identifies nearly twice as many variants capable of altering gene expression in cells compared to its predecessor.
The research team discovered that AlphaGenome 2 can also forecast outcomes for unfamiliar cell types and biological states without requiring retraining. Additionally, it can pinpoint gaps in experimental data, helping scientists determine their next laboratory investigations.
Adam Gayoso, a research scientist at Google DeepMind working on the initiative, told The Verge that the primary breakthrough lies in connecting human genetics directly to cell biology. The tool enables researchers to examine genetic variants found in large populations, observe their cellular-level effects, and potentially link those effects to specific diseases.
Applications and Future Plans
Žiga Avsec, genomics lead at Google DeepMind, told The Verge that such specificity is essential when working toward treatments. Developing drugs involves targeting particular proteins and cell types, making this granular data crucial, according to Avsec.
Avsec noted that creating a model like AlphaGenome 2 became feasible recently due to advanced technologies capable of capturing individual cells and their states at high resolution. DeepMind is currently working to enhance its genetics models by generating more data, particularly for underrepresented cell states. Future iterations might also be trained on data from multiple species to provide a broader biological view, while long-term goals include using models like AlphaGenome 2 to build digital simulations known as virtual cells.
Although Avsec considers AlphaGenome 2 a successor to the original AlphaGenome, it will not replace the older model immediately. The detailed predictions require greater computational resources and carry higher costs, meaning it would have been impractical to use AlphaGenome 2 for generating the genome-wide atlas released by DeepMind in September.
Google stated that AlphaGenome 2 is available immediately for non-commercial research through the AlphaGenome API, alongside a limited private preview intended for commercial research.
